Transparent Qt-widget and Ogre - qt

we ’ve been looking into using transparent widgets as overlays for ogre. Although Qwidgets are transparent by default, overlaying them over an Ogre widget is different and a black screen arises. On the ogre forum (http://www.ogre3d.org/forums/viewtopic.php?f=2&t=42733) I found the following:
"Ok, to get my ideas straight I’m going to summarise the various approaches I have found to combining Ogre and Qt:
Create a window (not necessarily a QGLWidget) using Qt and pass its window handle to Ogre using the ‘externalWindowHandle’ flag. Ogre then creates a context and is fully responsible for it. This is basically what I did for the first demo I made. It has the advantage that it should support both OpenGL and Direct3D, and also should work accross all platforms. The big drawback is that Qt widgets can only be overlaid (not integrated)
so you don’t get the nice transparency.
An approach similar to (1) but with the difference that the frame buffer is explicitly read back from Ogre and used as a widget background. This would again work accross all render systems and platforms, and now Qt can perform proper compositing from transparency. However, it is potenially slow doe to the need to read back every frame and copy it to Qt. But this could make a good ‘fallback’ approach. I haven’t tried implementing it yet.
This is the ‘proper’ approach which we are really considering in this thread. It involves both Ogre and Qt owning OpenGL contexts and both rendering into the same widget. This widget is set as the background of a QGraphicsView/QGraphicsScene and other widgets can be overliad with proper transpaency effects. It only works with OpenGL and is (theorectically) cross platform. There are two variations:
Allow Ogre to create it’s own OpenGL context (passing the flag ‘externalGLControl’) and then retieve the context once createRenderWindow() has finished. Under windows there are the wglGetCurrentDC() and wglGetCurrentContext() for this – other platforms should have something similar.
Have Qt create a context on Ogre’s behalf and pass that to Ogre using the ‘externalGLContext’ flag. In this case Ogre basically does nothing except render into the context it it provided with – all the control is with Qt. I haven’t quite got this to work yet as I was having crashes in both Qt and Ogre."
Approach 1 is the one we followed for a project and as said above we’ve encountered the same problem with transparency. Approach 3 is possible, but we would be limiting ourselves to openGL. The code for that is available one page 3 of the forum-link.
We’ve also found a demo video but I can’t find how it was implemented, but it looks really cool: http://youtu.be/b_3xfFuwGOQ

You can now find an implementation of the second variant of your third proposed approach in an updated QmlOgre example at https://github.com/advancingu/QmlOgre The original example had a few issues that were fixed here.
Hope this still helps a few months after your question.

I couldn't find sample implementations of options 2 and 3, so I'll share mine here.
I've tested the implementations below with Qt 5.4.0 x86 and OGRE 1.10.0 x86 on Windows 10.
Disclaimer: This is sample code. I'm not recommending you to:
implement everything in header files;
use owning raw pointers - use smart pointers, instead.
Direct OpenGL backend
A possible implementation of option 3.2 is the following:
// OgreGLWidget.h
#pragma once
#include "Scene.h"
#include "OgreSceneManager.h"
#include "OgreRenderSystem.h"
#include "OgreRenderWindow.h"
#include "OgreRoot.h"
#include <QtWidgets/QOpenGLWidget>
#include <algorithm>
#include <cstdint>
class OgreGLWidget final
: public QOpenGLWidget
{
Q_OBJECT
public:
explicit OgreGLWidget(QWidget* parent = nullptr)
:
QOpenGLWidget(parent),
Root_("plugins" OGRE_BUILD_SUFFIX ".cfg", "ogre.cfg")
{ }
~OgreGLWidget() override
{
this->Scene_.clear();
if (this->SceneManager_)
this->Root_.destroySceneManager(this->SceneManager_);
if (this->RenderWindow_)
this->RenderWindow_->destroy();
}
protected:
void initializeGL() override
{
QOpenGLWidget::initializeGL();
if (auto* glRenderSystem = this->getGlRenderSystem())
{
this->Root_.setRenderSystem(glRenderSystem);
auto* renderWindow = this->Root_.initialise(false);
assert(renderWindow == nullptr && "The render window must be created afterwards");
}
else
{
throw std::runtime_error("Cannot find GL render system");
}
Ogre::NameValuePairList params;
params["externalGLControl"] = "true";
params["hidden"] = "true";
// Pass the current OpenGL context to OGRE
// - either like this
params["currentGLContext"] = "true";
// - or like this
//if (auto const* openGLContext = QOpenGLContext::currentContext())
//{
// auto const nativeHandle = openGLContext->nativeHandle();
// if (!nativeHandle.isNull() && nativeHandle.canConvert<QWGLNativeContext>())
// {
// auto nativeContext = nativeHandle.value<QWGLNativeContext>();
// auto nativeRenderingContextHandle = nativeContext.context();
// params["externalGLContext"] = Ogre::StringConverter::toString(reinterpret_cast<unsigned long>(nativeRenderingContextHandle));
// qDebug("OpenGLContext: nativeHandle=%p, windowHandle=%p", nativeRenderingContextHandle, nativeContext.window());
// }
//}
// Setup scene
this->RenderWindow_ = this->Root_.createRenderWindow("", this->width(), this->height(), false, &params);
this->SceneManager_ = this->Root_.createSceneManager(Ogre::SceneType::ST_GENERIC);
this->SceneManager_->setAmbientLight(Ogre::ColourValue::White);
this->Scene_ = Scene(*this->SceneManager_, *this->RenderWindow_);
}
void paintGL() override
{
QOpenGLWidget::paintGL();
this->Root_.renderOneFrame();
// Queue another update to call paintGL again.
// Not ideal as it will render even if the scene hasn't changed.
this->update();
}
void resizeGL(int width, int height) override
{
QOpenGLWidget::resizeGL(width, height);
this->RenderWindow_->resize(width, height);
}
private:
Ogre::Root Root_;
Ogre::RenderWindow* RenderWindow_{ nullptr };
Ogre::SceneManager* SceneManager_{ nullptr };
Scene Scene_{};
Ogre::RenderSystem* getGlRenderSystem()
{
static Ogre::String const render_system_name{ "OpenGL Rendering Subsystem" };
auto const& renderSystems = this->Root_.getAvailableRenderers();
auto const lastRenderSystem = std::cend(renderSystems);
auto const glRenderSystemIt = std::find_if(std::cbegin(renderSystems), lastRenderSystem,
[](Ogre::RenderSystem const* renderSystem) { return renderSystem->getName() == render_system_name; });
return (glRenderSystemIt == lastRenderSystem) ? nullptr : *glRenderSystemIt;
}
};
Here you are a screenshot of the application running:
Buffered D3D9/OpenGL backend
A possible implementation of option 2 is the following:
// OgreWidget.h
#pragma once
#include "Scene.h"
#include "OgreRoot.h"
#include "OgreRenderWindow.h"
#include "OgreSceneManager.h"
#include <QtGui/qevent.h>
#include <QtGui/QPainter>
#include <QtWidgets/QWidget>
#include <cstdint>
class OgreWidget final
: public QWidget
{
Q_OBJECT
public:
explicit OgreWidget(QWidget *parent = nullptr)
:
QWidget(parent),
Root_("plugins" OGRE_BUILD_SUFFIX ".cfg", "ogre.cfg")
{
//static Ogre::String const render_system_name{ "OpenGL Rendering Subsystem" };
static Ogre::String const render_system_name{ "Direct3D9 Rendering Subsystem" };
if (auto* renderSystem = this->getRenderSystem(render_system_name))
{
this->Root_.setRenderSystem(renderSystem);
auto* renderWindow = this->Root_.initialise(false);
assert(renderWindow == nullptr && "The render window must be created afterwards");
}
else
{
throw std::runtime_error("Cannot find render system: " + render_system_name);
}
Ogre::NameValuePairList params;
params["hidden"] = "true";
this->RenderWindow_ = this->Root_.createRenderWindow("", this->width(), this->height(), false, &params);
this->SceneManager_ = this->Root_.createSceneManager(Ogre::SceneType::ST_GENERIC);
this->SceneManager_->setAmbientLight(Ogre::ColourValue::White);
this->Scene_ = Scene(*this->SceneManager_, *this->RenderWindow_);
}
~OgreWidget() override
{
this->Scene_.clear();
if (this->SceneManager_)
this->Root_.destroySceneManager(this->SceneManager_);
if (this->RenderWindow_)
this->RenderWindow_->destroy();
}
protected:
void paintEvent(QPaintEvent* evt) override
{
this->Root_.renderOneFrame();
this->RenderWindow_->update();
{
static auto const memory_category = Ogre::MemoryCategory::MEMCATEGORY_RENDERSYS;
static auto const pixel_format = Ogre::PixelFormat::PF_A8R8G8B8;
auto const width = this->RenderWindow_->getWidth();
auto const height = this->RenderWindow_->getHeight();
auto const bytesPerPixel = Ogre::PixelUtil::getNumElemBytes(pixel_format);
auto const byteCount = width * height * bytesPerPixel;
auto* data = OGRE_ALLOC_T(std::uint8_t, byteCount, memory_category);
Ogre::PixelBox pixelBox(width, height, 1, pixel_format, data);
this->RenderWindow_->copyContentsToMemory(pixelBox, pixelBox);
static auto const image_format = QImage::Format::Format_ARGB32;
QImage const background(data, width, height, image_format);
QPainter painter(this);
painter.drawImage(QPointF{ 0., 0. }, background, background.rect(), Qt::ImageConversionFlag::NoFormatConversion);
OGRE_FREE(data, memory_category);
}
QWidget::paintEvent(evt);
// Queue another update to call paintEvent again.
// Not ideal as it will render even if the scene hasn't changed.
this->update();
}
void resizeEvent(QResizeEvent* evt) override
{
auto const& size = evt->size();
auto const width = size.width();
auto const height = size.height();
this->RenderWindow_->resize(width, height);
QWidget::resizeEvent(evt);
}
private:
Ogre::Root Root_;
Ogre::RenderWindow* RenderWindow_{ nullptr };
Ogre::SceneManager* SceneManager_{ nullptr };
Scene Scene_;
Ogre::RenderSystem* getRenderSystem(Ogre::String const& renderSystemName)
{
auto const& renderSystems = this->Root_.getAvailableRenderers();
auto const lastRenderSystem = std::cend(renderSystems);
auto const renderSystemIt = std::find_if(std::cbegin(renderSystems), lastRenderSystem,
[&](Ogre::RenderSystem const* renderSystem) { return renderSystem->getName() == renderSystemName; });
return (renderSystemIt == lastRenderSystem) ? nullptr : *renderSystemIt;
}
};
Here you are a screenshot of the application running:
Performance
This approach is considerably slower than the OpenGL one.
On my machine, when the application is maximized, a frame is rendered in about:
~14ms with the (buffered) OpenGL backend;
~11ms with the (buffered) D3D9 backend;
~2 ms with the OpenGL backend;
Issues
For some reason, when using the buffered OpenGL backend, the cube has a different color:
I've not bothered to investigate it: if OpenGL is available, you most probably want to render the scene with class OgreGLWidget.
Boilerplate code
I'm embedding the OGRE scene as a child widget of centralWidget in my MainWindow instance:
// MainWindow.h
#pragma once
#include "OgreGLWidget.h"
#include "OgreWidget.h"
#include <QtWidgets/QLabel>
#include <QtWidgets/QLayout>
#include <QtWidgets/QMainWindow>
#include <QtWidgets/QPushButton>
#include "ui_MainWindow.h"
class MainWindow final
: public QMainWindow
{
Q_OBJECT
public:
explicit MainWindow(QWidget* parent = nullptr)
: QMainWindow(parent)
{
this->Ui_.setupUi(this);
auto* gridLayout = new QGridLayout(this->Ui_.centralWidget);
gridLayout->setSpacing(6);
gridLayout->setContentsMargins(0, 0, 0, 0);
// Uncomment based on whether you want to use OpenGL or D3D9 as backend.
this->OgreWidget_ = new OgreGLWidget(this->Ui_.centralWidget);
//this->OgreWidget_ = new OgreWidget(this->Ui_.centralWidget);
gridLayout->addWidget(this->OgreWidget_);
// Show an overlay widget when "Show Widget" is pressed.
auto* action = new QAction("Show Widget", this);
QObject::connect(action, &QAction::triggered, this, &MainWindow::showWidgetSlot);
this->Ui_.mainToolBar->addAction(action);
}
~MainWindow() override = default;
private slots:
void showWidgetSlot()
{
auto* w = new QWidget(this->OgreWidget_);
w->setAttribute(Qt::WidgetAttribute::WA_DeleteOnClose);
w->setAttribute(Qt::WidgetAttribute::WA_NoMousePropagation);
w->setStyleSheet(".QWidget { background-color: rgba(255,255,255,75%); border-radius: 10px; }");
auto layout = new QVBoxLayout(w);
layout->addWidget(new QLabel(this->tr("Hello OGRE"), w));
auto* b = new QPushButton(this->tr("Ok"), w);
layout->addWidget(b);
QObject::connect(b, &QPushButton::clicked, w, &QWidget::close);
w->resize(100, 100);
w->show();
}
private:
Ui::MainWindowClass Ui_;
QWidget* OgreWidget_{ nullptr };
};
Class Scene, responsible for setting up my scene, is implemented like this:
// Scene.h
#pragma once
#include "OgreCamera.h"
#include "OgreEntity.h"
#include "OgreMaterialManager.h"
#include "OgrePass.h"
#include "OgreRenderTarget.h"
#include "OgreSceneManager.h"
#include "OgreSceneNode.h"
#include "OgreTechnique.h"
#include "OgreViewport.h"
struct Scene
{
explicit Scene() = default;
explicit Scene(Ogre::SceneManager& sceneManager_, Ogre::RenderTarget& renderTarget_)
: sceneManager(&sceneManager_)
{
auto cubeMaterial = Ogre::MaterialManager::getSingleton().create("cubeMaterial", Ogre::ResourceGroupManager::DEFAULT_RESOURCE_GROUP_NAME);
if (auto* technique = cubeMaterial->getTechnique(0))
{
if (auto* pass = technique->getPass(0))
{
pass->setLightingEnabled(true);
pass->setAmbient(Ogre::ColourValue{ 0.3f, 0.1f, 0.1f, 1 });
}
}
this->entity = sceneManager_.createEntity(Ogre::SceneManager::PrefabType::PT_CUBE);
this->entity->setMaterial(cubeMaterial);
sceneManager_.getRootSceneNode()->attachObject(this->entity);
this->camera = sceneManager_.createCamera("camera");
this->camera->setPosition(-100, 100, 300);
this->camera->lookAt(0, 0, 0);
this->camera->setNearClipDistance(5);
this->light = sceneManager_.createLight("light");
this->light->setType(Ogre::Light::LightTypes::LT_DIRECTIONAL);
this->light->setDirection(0.25, -1, -0.5);
this->viewport = renderTarget_.addViewport(this->camera);
this->viewport->setBackgroundColour(Ogre::ColourValue::Blue);
}
Scene(Scene const&) = delete;
Scene(Scene&& other)
{
this->swap(other);
}
Scene& operator=(Scene const&) = delete;
Scene& operator=(Scene&& other)
{
this->swap(other);
return *this;
}
void swap(Scene& other)
{
using std::swap;
swap(this->sceneManager, other.sceneManager);
swap(this->viewport, other.viewport);
swap(this->entity, other.entity);
swap(this->camera, other.camera);
swap(this->light, other.light);
}
void clear()
{
if (this->light)
{
this->sceneManager->destroyLight(this->light);
this->light = nullptr;
}
if (this->camera)
{
this->sceneManager->destroyCamera(this->camera);
this->camera = nullptr;
}
if (this->entity)
{
this->sceneManager->destroyEntity(this->entity);
this->entity = nullptr;
}
this->sceneManager = nullptr;
}
~Scene()
{
this->clear();
}
// -- Non-owning
Ogre::SceneManager* sceneManager{ nullptr };
Ogre::Viewport* viewport{ nullptr };
// -- Owning
Ogre::Entity* entity{ nullptr };
Ogre::Camera* camera{ nullptr };
Ogre::Light* light{ nullptr };
};

*Here is the code for that video (http://youtu.be/b_3xfFuwGOQ). (Needs Qt5)
http://qt.gitorious.org/qt-labs/qmlogre
There is also Cutexture (Needs Qt4)
https://github.com/advancingu/Cutexture
http://i56.tinypic.com/b6xb83.jpg
Here is a Qt4 Ogre Widget example. (Needs Qt4)
http://zester.googlecode.com/files/QtCreatorOgre2.zip*

Related

How to render programmatically a vtk item in qml?

So far, I understand that we have two threads in QML, our main application thread, and our "scene graph" thread : http://doc.qt.io/qt-5/qtquick-visualcanvas-scenegraph.html
I've implemented my own vtkQmlItem with the help of this link : http://doc.qt.io/qt-5/qtquick-scenegraph-openglunderqml-example.html
and I've noticed that my vtkscene is only rendered when the afterrendering signal is emitted by the qml flow.
So far, everything is ok and works perfectly, I can see my vtk scene and can even interract with it.
But I would like to also programmatically render my vtk scene as well, since I want to do an animation by moving the camera around a vtk object.
Calling renderer->render() directly shows a lot of vtk error, and does not seem to be the good way to do this.
Calling this->window()->update() seems to put the event in the eventLoop, when I want it to be handled instantly. The only way I've managed to make it work instantly is by using QApplication::processEvents(), which is a hack I don't like and would love another solution.
So the pseudocode of the working solution that I don't like is the following :
for (int i = 0; i < 50; i++)
{
ChangeCameraPosition(i); // Change the position and orientation of the vtk camera
this->window()->update();
QApplication::processEvents(); // The hack I don't like
QThread::msleep(500);
}
For people looking for a solution for this using Qt QuickControls 2 and VTK 8, you can find a working example in this repository https://github.com/nicanor-romero/QtVtk with building instructions in the README.
the problem is actually a bit complicated and if nothing changed in the past few months, there is still no support for QtQuick in VTK, which means no simple few lines solution. You can find support classes for QtWidgets in VTK/GUISupport/QtOpenGL/ and use them as a template to derive support for qml. But mainly I recommend checking this thread for a discussion about this topic.
The key point is that QtQuick holds the openGL context for the qml window you are trying to render into in a dedicated thread and it won't let anything else get that context. So in order to render into it from VTK, you have to do it within that thread. This means:
1) Create your own vtkRenderWindow that overrides the Render() method such that it is ensured it happens in the qml's render thread.
2) Make that render window render into a framebuffer object provided by the qtquick (instance of QQuickFramebufferObject).
3) Interconnect vtk's rendering signals with the qt's rendering methods -> e.g. when the vtk render window calls makeCurrent, the qt's rendering thread "wakes up".
Here is my implementation based on Taylor Braun-Jones' template linked above. It might not be perfect, but it works for me (I have removed some parts specific to my app so it might not compile straight away, but it should put you on a path to some working solution):
qmlVtk.h:
#include <vtkEventQtSlotConnect.h>
#include <vtkGenericOpenGLRenderWindow.h>
#include <vtkRenderer.h>
#include <QtQuick/QQuickFramebufferObject>
// Use the OpenGL API abstraction from Qt instead of from VTK because vtkgl.h
// and other Qt OpenGL-related headers do not play nice when included in the
// same compilation unit
#include <QOpenGLFunctions>
#include <qqmlapplicationengine.h>
class QVTKFramebufferObjectRenderer;
class QVTKInteractorAdapter;
class vtkInternalOpenGLRenderWindow;
class QVTKFramebufferObjectRenderer;
class QVTKFrameBufferObjectItem : public QQuickFramebufferObject
{
Q_OBJECT
public:
QVTKFrameBufferObjectItem(QQuickItem *parent = 0);
~QVTKFrameBufferObjectItem();
Renderer *createRenderer() const;
vtkSmartPointer<vtkInternalOpenGLRenderWindow> GetRenderWindow() const;
protected:
// Called once before the FBO is created for the first time. This method is
// called from render thread while the GUI thread is blocked.
virtual void init();
vtkSmartPointer<vtkInternalOpenGLRenderWindow> m_win;
QVTKInteractorAdapter* m_irenAdapter;
vtkSmartPointer<vtkEventQtSlotConnect> mConnect;
friend class QVTKFramebufferObjectRenderer;
// Convert the position of the event from openGL coordinate to native coordinate
QMouseEvent openGLToNative(QMouseEvent const& event);
virtual void mouseMoveEvent(QMouseEvent * event);
virtual void mousePressEvent(QMouseEvent * event);
virtual void mouseReleaseEvent(QMouseEvent * event);
virtual void mouseDoubleClickEvent(QMouseEvent * event);
virtual void wheelEvent(QWheelEvent *event);
virtual void keyPressEvent(QKeyEvent* event);
virtual void keyReleaseEvent(QKeyEvent* event);
virtual void focusInEvent(QFocusEvent * event);
virtual void focusOutEvent(QFocusEvent * event);
protected Q_SLOTS:
// slot to make this vtk render window current
virtual void MakeCurrent();
// slot called when vtk wants to know if the context is current
virtual void IsCurrent(vtkObject* caller, unsigned long vtk_event, void* client_data, void* call_data);
// slot called when vtk wants to start the render
virtual void Start();
// slot called when vtk wants to end the render
virtual void End();
// slot called when vtk wants to know if a window is direct
virtual void IsDirect(vtkObject* caller, unsigned long vtk_event, void* client_data, void* call_data);
// slot called when vtk wants to know if a window supports OpenGL
virtual void SupportsOpenGL(vtkObject* caller, unsigned long vtk_event, void* client_data, void* call_data);
};
/// <summary>
/// An extension of vktGenericOpenGLRenderWindow to work with Qt.
/// Serves to write VTK-generated render calls to a framebuffer provided and maintained by Qt. It is meant to be used within Qt render loop, i.e. using Qt's render thread.
/// </summary>
/// <seealso cref="vtkGenericOpenGLRenderWindow" />
/// <seealso cref="QOpenGLFunctions" />
class vtkInternalOpenGLRenderWindow : public vtkGenericOpenGLRenderWindow, protected QOpenGLFunctions
{
public:
static vtkInternalOpenGLRenderWindow* New();
vtkTypeMacro(vtkInternalOpenGLRenderWindow, vtkGenericOpenGLRenderWindow)
virtual void OpenGLInitState();
// Override to use deferred rendering - Tell the QSG that we need to
// be rendered which will then, at the appropriate time, call
// InternalRender to do the actual OpenGL rendering.
virtual void Render();
// Do the actual OpenGL rendering
void InternalRender();
// Provides a convenient way to set the protected FBO ivars from an existing
// FBO that was created and owned by Qt's FBO abstraction class
// QOpenGLFramebufferObject
void SetFramebufferObject(QOpenGLFramebufferObject *fbo);
QVTKFramebufferObjectRenderer *QtParentRenderer;
protected:
vtkInternalOpenGLRenderWindow();
~vtkInternalOpenGLRenderWindow()
{
// Prevent superclass destructors from destroying the framebuffer object.
// QOpenGLFramebufferObject owns the FBO and manages it's lifecyle.
this->OffScreenRendering = 0;
}
};
qmlVtk.cpp:
#include "QVTKFramebufferObjectItem.h"
#include <QQuickFramebufferObject>
#include <QQuickWindow>
#include <QOpenGLFramebufferObject>
#include <QVTKInteractorAdapter.h>
#include <vtkRenderWindowInteractor.h>
#include <vtkObjectFactory.h>
#include <vtkSmartPointer.h>
#include <vtkCamera.h>
#include <vtkProperty.h>
#include <qglfunctions.h>
class QVTKFramebufferObjectRenderer : public QQuickFramebufferObject::Renderer
{
friend class vtkInternalOpenGLRenderWindow;
public:
QVTKFramebufferObjectRenderer(vtkSmartPointer<vtkInternalOpenGLRenderWindow> rw) :
m_framebufferObject(0)
{
m_vtkRenderWindow = rw;
m_vtkRenderWindow->QtParentRenderer = this;
}
~QVTKFramebufferObjectRenderer()
{
m_vtkRenderWindow->QtParentRenderer = 0;
glFrontFace(GL_CCW); // restore default settings
}
virtual void synchronize(QQuickFramebufferObject * item)
{
// the first synchronize call - right before the the framebufferObject
// is created for the first time
if (!m_framebufferObject)
{
QVTKFrameBufferObjectItem *vtkItem = static_cast<QVTKFrameBufferObjectItem*>(item);
vtkItem->init();
}
}
virtual void render()
{
m_vtkRenderWindow->InternalRender(); // vtkXOpenGLRenderWindow renders the scene to the FBO
}
QOpenGLFramebufferObject *createFramebufferObject(const QSize &size)
{
QOpenGLFramebufferObjectFormat format;
format.setAttachment(QOpenGLFramebufferObject::Depth);
m_framebufferObject = new QOpenGLFramebufferObject(size, format);
m_vtkRenderWindow->SetFramebufferObject(m_framebufferObject);
return m_framebufferObject;
}
vtkSmartPointer<vtkInternalOpenGLRenderWindow> m_vtkRenderWindow;
QOpenGLFramebufferObject *m_framebufferObject;
};
vtkStandardNewMacro(vtkInternalOpenGLRenderWindow);
vtkInternalOpenGLRenderWindow::vtkInternalOpenGLRenderWindow() :
QtParentRenderer(0)
{
vtkOpenGLRenderWindow::OpenGLInitContext();
}
void vtkInternalOpenGLRenderWindow::OpenGLInitState()
{
this->MakeCurrent();
vtkOpenGLRenderWindow::OpenGLInitState();
// Before any of the gl* functions in QOpenGLFunctions are called for a
// given OpenGL context, an initialization must be run within that context
initializeOpenGLFunctions();
glFrontFace(GL_CW); // to compensate for the switched Y axis
}
void vtkInternalOpenGLRenderWindow::InternalRender()
{
vtkOpenGLRenderWindow::Render();
}
//
// vtkInternalOpenGLRenderWindow Definitions
//
void vtkInternalOpenGLRenderWindow::Render()
{
this->QtParentRenderer->update();
}
void vtkInternalOpenGLRenderWindow::SetFramebufferObject(QOpenGLFramebufferObject *fbo)
{
// QOpenGLFramebufferObject documentation states that "The color render
// buffer or texture will have the specified internal format, and will
// be bound to the GL_COLOR_ATTACHMENT0 attachment in the framebuffer
// object"
this->BackLeftBuffer = this->FrontLeftBuffer = this->BackBuffer = this->FrontBuffer =
static_cast<unsigned int>(GL_COLOR_ATTACHMENT0);
// Save GL objects by static casting to standard C types. GL* types
// are not allowed in VTK header files.
QSize fboSize = fbo->size();
this->Size[0] = fboSize.width();
this->Size[1] = fboSize.height();
this->NumberOfFrameBuffers = 1;
this->FrameBufferObject = static_cast<unsigned int>(fbo->handle());
this->DepthRenderBufferObject = 0; // static_cast<unsigned int>(depthRenderBufferObject);
this->TextureObjects[0] = static_cast<unsigned int>(fbo->texture());
this->OffScreenRendering = 1;
this->OffScreenUseFrameBuffer = 1;
this->Modified();
}
void QVTKFrameBufferObjectItem::Start()
{
m_win->OpenGLInitState();
}
void QVTKFrameBufferObjectItem::End()
{
}
void QVTKFrameBufferObjectItem::MakeCurrent()
{
this->window()->openglContext()->makeCurrent(this->window());
}
void QVTKFrameBufferObjectItem::IsCurrent(vtkObject*, unsigned long, void*, void* call_data)
{
bool* ptr = reinterpret_cast<bool*>(call_data);
*ptr = this->window()->openglContext();
}
void QVTKFrameBufferObjectItem::IsDirect(vtkObject*, unsigned long, void*, void* call_data)
{
int* ptr = reinterpret_cast<int*>(call_data);
*ptr = QGLFormat::fromSurfaceFormat(this->window()->openglContext()->format()).directRendering();
}
void QVTKFrameBufferObjectItem::SupportsOpenGL(vtkObject*, unsigned long, void*, void* call_data)
{
int* ptr = reinterpret_cast<int*>(call_data);
*ptr = QGLFormat::hasOpenGL();
}
QVTKFrameBufferObjectItem::QVTKFrameBufferObjectItem(QQuickItem *parent) : QQuickFramebufferObject(parent)
{
setAcceptedMouseButtons(Qt::AllButtons);
m_irenAdapter = new QVTKInteractorAdapter(this);
m_win = vtkSmartPointer<vtkInternalOpenGLRenderWindow>::New();
// make a connection between the vtk signals and qt slots so that an initialized and madeCurrent opengl context is given to the vtk
// we probably need only the Start(), MakeCurrent() and End() one, but just to be sure...
mConnect = vtkSmartPointer<vtkEventQtSlotConnect>::New();
mConnect->Connect(m_win, vtkCommand::WindowMakeCurrentEvent, this, SLOT(MakeCurrent()));
mConnect->Connect(m_win, vtkCommand::WindowIsCurrentEvent, this, SLOT(IsCurrent(vtkObject*, unsigned long, void*, void*)));
mConnect->Connect(m_win, vtkCommand::StartEvent, this, SLOT(Start()));
mConnect->Connect(m_win, vtkCommand::EndEvent, this, SLOT(End()));
mConnect->Connect(m_win, vtkCommand::WindowIsDirectEvent, this, SLOT(IsDirect(vtkObject*, unsigned long, void*, void*)));
mConnect->Connect(m_win, vtkCommand::WindowSupportsOpenGLEvent, this, SLOT(SupportsOpenGL(vtkObject*, unsigned long, void*, void*)));
}
QVTKFrameBufferObjectItem::~QVTKFrameBufferObjectItem()
{
mConnect->Disconnect(); // disconnect all slots
if (m_irenAdapter)
delete m_irenAdapter;
}
QQuickFramebufferObject::Renderer *QVTKFrameBufferObjectItem::createRenderer() const
{
return new QVTKFramebufferObjectRenderer(m_win);
}
vtkSmartPointer<vtkInternalOpenGLRenderWindow> QVTKFrameBufferObjectItem::GetRenderWindow() const
{
return m_win;
}
void QVTKFrameBufferObjectItem::init()
{
}
// theoretically not needed now - the Y is being flipped in render and devicePixelRatio will almost always be = 1 on a PC anyway...but lets keep it to be sure
QMouseEvent QVTKFrameBufferObjectItem::openGLToNative(QMouseEvent const& event)
{
QPointF localPos(event.localPos());
localPos.setX(localPos.x() * window()->devicePixelRatio());
localPos.setY(localPos.y() * window()->devicePixelRatio());
QMouseEvent nativeEvent(event.type(), localPos, event.button(), event.buttons(), event.modifiers());
return nativeEvent;
}
void QVTKFrameBufferObjectItem::mouseMoveEvent(QMouseEvent * event)
{
m_win->GetInteractor()->SetSize(this->width(), this->height());
QMouseEvent nativeEvent = openGLToNative(*event);
m_irenAdapter->ProcessEvent(&nativeEvent, this->m_win->GetInteractor());
}
void QVTKFrameBufferObjectItem::mousePressEvent(QMouseEvent * event)
{
m_win->GetInteractor()->SetSize(this->width(), this->height());
QMouseEvent nativeEvent = openGLToNative(*event);
m_irenAdapter->ProcessEvent(&nativeEvent, this->m_win->GetInteractor());
}
void QVTKFrameBufferObjectItem::mouseReleaseEvent(QMouseEvent * event)
{
m_win->GetInteractor()->SetSize(this->width(), this->height());
QMouseEvent nativeEvent = openGLToNative(*event);
m_irenAdapter->ProcessEvent(&nativeEvent, this->m_win->GetInteractor());
}
void QVTKFrameBufferObjectItem::wheelEvent(QWheelEvent *event)
{
m_irenAdapter->ProcessEvent(event, this->m_win->GetInteractor());
}
void QVTKFrameBufferObjectItem::keyPressEvent(QKeyEvent* event)
{
m_irenAdapter->ProcessEvent(event, this->m_win->GetInteractor());
}
void QVTKFrameBufferObjectItem::keyReleaseEvent(QKeyEvent* event)
{
m_irenAdapter->ProcessEvent(event, this->m_win->GetInteractor());
}
void QVTKFrameBufferObjectItem::focusInEvent(QFocusEvent * event)
{
m_irenAdapter->ProcessEvent(event, this->m_win->GetInteractor());
}
void QVTKFrameBufferObjectItem::focusOutEvent(QFocusEvent * event)
{
m_irenAdapter->ProcessEvent(event, this->m_win->GetInteractor());
}
To use it, define an instance of the framebuffer in your qml form and stretch it across the window you want to render into, e.g. like this (assuming you registered the QVTKFrameBufferObjectItem as a QVTKFrameBuffer in qml e.g. like this qmlRegisterType<QVTKFrameBufferObjectItem>("VtkQuick", 1, 0, "QVTKFrameBuffer");):
import VtkQuick 1.0
QVTKFrameBuffer
{
id: renderBuffer
anchors.fill : parent
Component.onCompleted :
{
myCppDisplay.framebuffer = renderBuffer // tell the c++ side of your app that this is the framebuffer into which it should render
}
}
You then use the vtkRenderWindow you get by myCppDisplay.framebuffer.GetRenderWindow() the same way you would use any other vtkRenderWindow if you were rendering into a vtk-managed window, i.e. you can assign vtkRenderer to it, assign actors to that renderer, call theWindow.Render() as you wish and it will all be rendered into the qml component to which you assigned the framebuffer.
Two notes: 1) the vtk and qt use different coordinate system, you need to flip the y-coordinate...I am doing it by assigning a scale transformation to the camera, but there is plenty of other ways to do it:
vtkSmartPointer<vtkTransform> scale = vtkSmartPointer<vtkTransform>::New();
scale->Scale(1, -1, 1);
renderer->GetActiveCamera()->SetUserTransform(scale);
2) things get quite tricky once you start using multiple threads - you have to make sure that you are not trying to render in two different threads, because they would compete for that one QtQuick's rendering thread. This does not mean only not calling renderWindow.Render() in parallel - that is easy to avoid - but you have to realize that that qt thread is used also for rendering the GUI, so you might get into trouble this way (updating GUI while doing VTK rendering).
I tried implementing this example and have successfully compiled and ran it as well but I am facing some challenges with respect to the code.
I am not able to figure out where these lines are defined as they are giving me runtime errors and are not getting recognized as well.
(Defined in qmlVtk.cpp -> SetFramebufferObject() function)
this->NumberOfFrameBuffers = 1;
this->FrameBufferObject = static_cast<unsigned int>(fbo->handle());
this->DepthRenderBufferObject = 0;
this->TextureObjects[0] = static_cast<unsigned int>(fbo->texture());
this->OffScreenRendering = 1;
this->OffScreenUseFrameBuffer = 1;
Also, can anyone guide me as how to integrate QML with OpenGLWidget.

How to monitor changes to an arbitrary widget?

I am starting a QT5 application with a rather complex design based on Qt Widgets. It runs on Beagleboard with a touchscreen. I will have a rather weird local invention instead of the LCD display. It's a laser painting on acrylic plate. It has no driver yet. To actually update a screen I must create a screenshot of the window as bitmap, turn it to grayscale and feed to a proprietary library, which will handle the laser. It should look cute, when ready. Unfortunately, the laser blinks on update, so I cannot just make screenshots on timer, or it will be jerky like hell.
I need to run a function every time a meaningful update of GUI happens, while preferably ignore things like button being pressed and released. Is there some way to create a hook without subclassing every single Qt Widget I will use? The only way to do this I know is to override paintEvent of everything. I want a simpler solution.
Possible assumptions are: the application will be running under X server with dummy display, will be the only GUI app running. Some updates happen without user input.
The code below does it. It doesn't dig too deeply into the internals of Qt, it merely leverages the fact that backing store devices are usually QImages. It could be modified to accommodate OpenGL-based backing stores as well.
The WidgetMonitor class is used to monitor the widgets for content changes. An entire top-level window is monitored no matter which particular widget is passed to the monitor(QWidget*) method. You only need to call the monitor method for one widget in the window you intend to monitor - any widget will do. The changes are sent out as a QImage of window contents.
The implementation installs itself as an event filter in the target window widget and all of its children, and monitors the repaint events. It attempts to coalesce the repaint notifications by using the zero-length timer. The additions and removals of children are tracked automagically.
When you run the example, it creates two windows: a source window, and a destination window. They may be overlapped so you need to separate them. As you resize the source window, the size of the destination's rendition of it will also change appropriately. Any changes to the source children (time label, button state) propagate automatically to the destination.
In your application, the destination could be an object that takes the QImage contents, converts them to grayscale, resizes appropriately, and passes them to your device.
I do not quite understand how your laser device works if it can't gracefully handle updates. I presume that it is a raster-scanning laser that runs continuously in a loop that looks roughly like this:
while (1) {
for (line = 0; line < nLines; ++line) {
drawLine();
}
}
You need to modify this loop so that it works as follows:
newImage = true;
QImage localImage;
while (1) {
if (newImage) localImage = newImage;
for (line = 0; line < localImage.height(); ++line) {
drawLine(line, localImage);
}
}
You'd be flipping the newImage flag from the notification slot connected to the WidgetMonitor. You may well find out that leveraging QImage, and Qt's functionality in general, in your device driver code, will make it much easier to develop. Qt provides portable timers, threads, collections, etc. I presume that your "driver" is completely userspace, and communicates via a serial port or ethernet to the micro controller that actually controls the laser device.
If you will be writing a kernel driver for the laser device, then the interface would be probably very similar, except that you end up writing the image bitmap to an open device handle.
// https://github.com/KubaO/stackoverflown/tree/master/questions/surface-20737882
#include <QtWidgets>
#include <array>
const char kFiltered[] = "WidgetMonitor_filtered";
class WidgetMonitor : public QObject {
Q_OBJECT
QVector<QPointer<QWidget>> m_awake;
QBasicTimer m_timer;
int m_counter = 0;
void queue(QWidget *window) {
Q_ASSERT(window && window->isWindow());
if (!m_awake.contains(window)) m_awake << window;
if (!m_timer.isActive()) m_timer.start(0, this);
}
void filter(QObject *obj) {
if (obj->isWidgetType() && !obj->property(kFiltered).toBool()) {
obj->installEventFilter(this);
obj->setProperty(kFiltered, true);
}
}
void unfilter(QObject *obj) {
if (obj->isWidgetType() && obj->property(kFiltered).toBool()) {
obj->removeEventFilter(this);
obj->setProperty(kFiltered, false);
}
}
bool eventFilter(QObject *obj, QEvent *ev) override {
switch (ev->type()) {
case QEvent::Paint: {
if (!obj->isWidgetType()) break;
if (auto *window = static_cast<QWidget *>(obj)->window()) queue(window);
break;
}
case QEvent::ChildAdded: {
auto *cev = static_cast<QChildEvent *>(ev);
if (auto *child = qobject_cast<QWidget *>(cev->child())) monitor(child);
break;
}
default:
break;
}
return false;
}
void timerEvent(QTimerEvent *ev) override {
if (ev->timerId() != m_timer.timerId()) return;
qDebug() << "painting: " << m_counter++ << m_awake;
for (auto w : m_awake)
if (auto *img = dynamic_cast<QImage *>(w->backingStore()->paintDevice()))
emit newContents(*img, w);
m_awake.clear();
m_timer.stop();
}
public:
explicit WidgetMonitor(QObject *parent = nullptr) : QObject{parent} {}
explicit WidgetMonitor(QWidget *w, QObject *parent = nullptr) : QObject{parent} {
monitor(w);
}
Q_SLOT void monitor(QWidget *w) {
w = w->window();
if (!w) return;
filter(w);
for (auto *obj : w->findChildren<QWidget *>()) filter(obj);
queue(w);
}
Q_SLOT void unMonitor(QWidget *w) {
w = w->window();
if (!w) return;
unfilter(w);
for (auto *obj : w->findChildren<QWidget *>()) unfilter(obj);
m_awake.removeAll(w);
}
Q_SIGNAL void newContents(const QImage &, QWidget *w);
};
class TestWidget : public QWidget {
QVBoxLayout m_layout{this};
QLabel m_time;
QBasicTimer m_timer;
void timerEvent(QTimerEvent *ev) override {
if (ev->timerId() != m_timer.timerId()) return;
m_time.setText(QTime::currentTime().toString());
}
public:
explicit TestWidget(QWidget *parent = nullptr) : QWidget{parent} {
m_layout.addWidget(&m_time);
m_layout.addWidget(new QLabel{"Static Label"});
m_layout.addWidget(new QPushButton{"A Button"});
m_timer.start(1000, this);
}
};
int main(int argc, char **argv) {
QApplication app{argc, argv};
TestWidget src;
QLabel dst;
dst.setFrameShape(QFrame::Box);
for (auto *w : std::array<QWidget *, 2>{&dst, &src}) {
w->show();
w->raise();
}
QMetaObject::invokeMethod(&dst, [&] { dst.move(src.frameGeometry().topRight()); },
Qt::QueuedConnection);
WidgetMonitor mon(&src);
src.setWindowTitle("Source");
dst.setWindowTitle("Destination");
QObject::connect(&mon, &WidgetMonitor::newContents, [&](const QImage &img) {
dst.resize(img.size());
dst.setPixmap(QPixmap::fromImage(img));
});
return app.exec();
}
#include "main.moc"

Make a floating QDockWidget unfocusable

I'm creating a simple virtual keyboard in a QDockWidget...
When the widget is docked into the QMainWindow, the selected widget (for example a qdoublespinbox) is highlighted and if I click on the virtual keyboard clearFocus() works...
When the QDockWidget is floating above the window and I click a button, clearFocus doesn't work and I can't see the focused widget in QMainWindow...
How can I force the QDockWidget to not have any focus at all?
Thanks :-)
This is the code:
// class MyVirtualKeyboard : public QDockWidget
void MyVirtualKeyboard::sendKey(Qt::Key key, Qt::KeyboardModifier mod)
{
this->clearFocus();
QMainWindow *w = dynamic_cast<QMainWindow *>(this->parent());
if(w == NULL) return;
QWidget *widget = w->focusWidget();
QString repr = QKeySequence(key).toString();
QKeyEvent *pressEvent = new QKeyEvent(QEvent::KeyPress, key, mod, repr);
QKeyEvent *releaseEvent = new QKeyEvent(QEvent::KeyRelease, key, mod, repr);
qDebug("%s", pressEvent->text().toAscii().data());
MyApplication *app = MyApplication::myInstance();
app->postEvent(widget, pressEvent);
app->postEvent(widget, releaseEvent);
}
void MyVirtualKeyboard::on_BTN_1_clicked()
{
sendKey(Qt::Key_1);
}
...
The clearFocus() call should be unnecessary. Your dock widget and all of its widgets must have the Qt::NoFocus policy.
The code below shows how you might do it.
// https://github.com/KubaO/stackoverflown/tree/master/questions/vkb-focus-18558664
#include <QtGui>
#if QT_VERSION >= QT_VERSION_CHECK(5,0,0)
#include <QtWidgets>
#endif
class Keyboard : public QDockWidget {
Q_OBJECT
QWidget m_widget;
QGridLayout m_layout{&m_widget};
QToolButton m_buttons[10];
void sendKey(Qt::Key key, Qt::KeyboardModifier mod)
{
if (! parentWidget()) return;
auto target = parentWidget()->focusWidget();
if (! target) return;
auto repr = QKeySequence(key).toString();
auto pressEvent = new QKeyEvent(QEvent::KeyPress, key, mod, repr);
auto releaseEvent = new QKeyEvent(QEvent::KeyRelease, key, mod, repr);
qApp->postEvent(target, pressEvent);
qApp->postEvent(target, releaseEvent);
qDebug() << repr;
}
Q_SLOT void clicked() {
auto key = sender()->property("key");
if (key.isValid()) sendKey((Qt::Key)key.toInt(), Qt::NoModifier);
}
public:
explicit Keyboard(const QString & title, QWidget *parent = nullptr) : Keyboard(parent) {
setWindowTitle(title);
}
explicit Keyboard(QWidget *parent = nullptr) : QDockWidget(parent) {
int i{};
for (auto & btn : m_buttons) {
btn.setText(QString::number(i));
btn.setProperty("key", Qt::Key_0 + i);
m_layout.addWidget(&btn, 0, i, 1, 1);
connect(&btn, SIGNAL(clicked()), SLOT(clicked()));
btn.setFocusPolicy(Qt::NoFocus);
++i;
}
setWidget(&m_widget);
setFeatures(QDockWidget::DockWidgetFloatable | QDockWidget::DockWidgetMovable);
}
};
int main(int argc, char ** argv)
{
QApplication a(argc, argv);
QMainWindow w;
w.setCentralWidget(new QLineEdit);
w.addDockWidget(Qt::TopDockWidgetArea, new Keyboard("Keyboard", &w));
w.show();
return a.exec();
}
#include "main.moc"
You can prevent a widget from taking focus by setting QWidget::focusPolicy = Qt::NoFocus.
However, there are two concepts here that you're mixing - the focused control (per window), and the active window (per desktop). I think in the scenario you're describing (a torn-off popup window), the OS window manager is likely to still change the active top-level window even if Qt doesn't set a focused control. That will result in nobody having keyboard focus (which is a valid state!).
So I think a full answer to your question will involve some non-portable bits. I don't know what GUI environment you're working in, but I know some of the answer for Win32, so I'll keep going and hope that's useful:
Win32
There's a pretty good discussion of the state tracking for Win32 on MSDN in the article Win32 Activation and Focus. I'm not aware that Qt does anything to wrap this level, so you'd have to use QWidget::nativeEvent or QCoreApplication::installNativeEventFilter to get at the low-level event. If you can subclass the window, I'd prefer the former, since it's more self-contained.
bool FooWidget::nativeEvent(const QByteArray & eventType, void * message, long * result)
{
#ifdef Q_OS_WIN
if(eventType == "windows_generic_MSG") {
const MSG *msg = reinterpret_cast<MSG *>(message);
if(msg->message == WM_MOUSEACTIVATE) {
*result = MA_NOACTIVATE;
return true;
}
}
#else
#error Need platform-specific code to suppress click-activation
#endif
return false;
}
This should block the click from activating the window (MA_NOACTIVATE), and block Qt from processing it further (return true), while leaving other all events (including the the click, since we didn't use MA_NOACTIVATEANDEAT to block it too) to be processed into QEvents and Qt signals normally (return false at the end).
If you need further low-level access (though I don't think you will), see also QWidget::effectiveWinId() and QWidget::windowHandle
Thanks a lot to Martin Gräßlin for the answer!
My recommendation: check out the virtual keyboard code in KDE Plasma: http://quickgit.kde.org/?p=kdeplasma-addons.git&a=blob&h=5628d6325afe57f85917dad865a07d4116335726&hb=a658d1e257cfca2a43c12714d026ec26f1fdb755&f=applets%2Fplasmaboard%2Fwidget.cpp
Looks like the key is setWindowFlags(Qt::X11BypassWindowManagerHint) and setFocusPolicy(Qt::NoFocus)
MyVirtualKeyboard::MyVirtualKeyboard(QWidget *parent) :
QDockWidget(parent),
ui(new Ui::MyVirtualKeyboard)
{
ui->setupUi(this);
this->connect(this, SIGNAL(topLevelChanged(bool)), this, SLOT(topLevelChanged()));
}
void MyVirtualKeyboard::topLevelChanged()
{
if(this->isWindow())
{
this->setWindowFlags(Qt::Popup | Qt::WindowStaysOnTopHint | Qt::X11BypassWindowManagerHint);
this->setFocusPolicy(Qt::NoFocus);
this->show();
}
}
I think I've found a better way to do it!
Just use this->setAttribute(Qt::WA_X11DoNotAcceptFocus); and voila!
Example:
MyVirtualKeyboard::MyVirtualKeyboard(QWidget *parent) :
QDockWidget(parent),
ui(new Ui::MyVirtualKeyboard)
{
ui->setupUi(this);
this->setAttribute(Qt::WA_X11DoNotAcceptFocus);
}

Sliding OSX Dock type "sidebar" rather embedded QDockWidget type?

I want to make a sliding "sidebar" similar to the functionality of the OSX "Dock" (e.g. mouse passes edge of screen and Dock slides out). I've been playing around with QDockWidget but since that is embedded in the window layout, it causes everything to shift when it becomes visible.
Can someone suggest a way to implement this?
Doesn't need to float (as a separate window/tool bar)
Should scale to window height (e.g. window can be fullscreen or default size)
Doesn't need to slide (animate) if that is complicated.
I'm new to Qt and so don't want to over-think this. Is this just a matter of a custom widget or should I be looking at a borderless window? The custom widget approach seems right but I don't know how to specify that it overlay other window content and also scale if the window scales.
QDockWidget has nothing to do with what you want - behaviorally. Just because it's called a Dock widget doesn't mean it's the same "Dock" concept as in OS X. It merely means that it docks somewhere. QDockWidget's documentation quite explicitly explains what is meant by the docking behavior.
The code below implements the behavior you seem to want. Whether it's good design or not is arguable. The reason the code is "convoluted" seems to hint that nobody is expected to come up with such a UI design. What's wrong with actually clicking a button somewhere to display the slider window?
The code works under both Qt 4.8 and 5.1.
Note: This begs to be implemented in Qt Quick 2. That's what it was designed for :) Of course Qt 4.6+ improved the behavior of the QWidget-moving animations, and Qt 5 does further tweaks, but really this code smells bad and there's a good reason it does: QWidget API, while powerful, ultimately encapsulates a set of APIs that date to 1984 when the original Macintosh was released. There's only so much you can do when you have to composite results from a bunch of stacked painters. In Qt Quick, the rendering is done by the GPU. The animation amounts to passing a couple of new floats to the GPU to update a single transformation matrix. That's it.
#include <QApplication>
#include <QWidget>
#include <QGridLayout>
#include <QLabel>
#include <QPainter>
#include <QGradient>
#include <QMouseEvent>
#include <QPropertyAnimation>
class Slider : public QWidget {
void paintEvent(QPaintEvent *) Q_DECL_OVERRIDE {
QPainter p(this);
QLinearGradient g(QPointF(0,0), QPointF(rect().bottomRight()));
g.setColorAt(0, Qt::blue);
g.setColorAt(1, Qt::gray);
p.setBackground(g);
p.eraseRect(rect());
p.setPen(Qt::yellow);
p.setFont(QFont("Helvetica", 48));
p.drawText(rect(), "Click Me To Hide");
}
void mousePressEvent(QMouseEvent *) Q_DECL_OVERRIDE {
hide();
}
public:
explicit Slider(QWidget *parent = 0) : QWidget(parent) {
setAttribute(Qt::WA_OpaquePaintEvent);
}
};
class Window : public QWidget {
QGridLayout m_layout;
Slider m_slider;
QLabel m_label;
QPropertyAnimation m_animation;
public:
explicit Window(QWidget *parent = 0, Qt::WindowFlags f = 0) :
QWidget(parent, f),
m_layout(this),
m_slider(this),
m_animation(&m_slider, "pos")
{
setMouseTracking(true);
m_layout.addWidget(&m_label);
m_slider.hide();
m_slider.setMouseTracking(false);
m_animation.setStartValue(QPoint(-width(), 0));
m_animation.setEndValue(QPoint(0, 0));
m_animation.setDuration(500);
m_animation.setEasingCurve(QEasingCurve::InCubic);
}
void leaveEvent(QEvent *) {
if (window() && QCursor::pos().x() <= window()->geometry().topLeft().x()) {
showSlider();
}
}
void childEvent(QChildEvent * ev) {
if (ev->added() && ev->child()->isWidgetType()) {
ev->child()->installEventFilter(this);
static_cast<QWidget*>(ev->child())->setMouseTracking(true);
}
}
bool event(QEvent * ev) {
eventFilter(this, ev);
return QWidget::event(ev);
}
bool eventFilter(QObject *, QEvent * ev) {
if (ev->type() == QEvent::MouseMove) {
auto pos = QCursor::pos();
if (window() && window()->isFullScreen()) {
if (pos.x() <= window()->geometry().topLeft().x()) {
showSlider();
}
}
m_label.setText(QString("%1, %2").arg(pos.x()).arg(pos.y()));
}
return false;
}
void resizeEvent(QResizeEvent *) {
m_slider.resize(size());
m_animation.setStartValue(QPoint(-width(), 0));
}
Q_SLOT void showSlider() {
if (m_slider.isVisible() || (window() && qApp->activeWindow() != window())) return;
m_slider.raise();
m_slider.show();
m_animation.start();
}
};
int main(int argc, char *argv[])
{
QApplication a(argc, argv);
Window w;
w.show();
return a.exec();
}

How to change a parent widget's background when a child widget has focus?

I would like to highlight a QFrame, if one of it's child widgets has focus (so the users know where to look for the cursor ;-)
using something along
ui->frame->setFocusPolicy(Qt::StrongFocus);
ui->frame->setStyleSheet("QFrame:focus {background-color: #FFFFCC;}");
highlights the QFrame when I click on it, but it loses its focus once one of its child widgets is selected.
Possible approaches:
I could connect() QApplication::focusChanged(old,now) and check each new object if it is a child of my QFrame, but this gets messy.
I could also subclass each child widget and reimplement focusInEvent()/focusOutEvent() and react on that, but with a lot of different widgets, this is also a lot of work.
Is there a more elegant solution?
Well, you can extend QFrame to make it listen on focus change of its children widgets.
Or you can also install an event filter on children widgets to catch QFocusEvent.
Here is an example:
MyFrame.h
#ifndef MYFRAME_H
#define MYFRAME_H
#include <QFrame>
class MyFrame : public QFrame
{
Q_OBJECT
public:
explicit MyFrame(QWidget* parent = 0, Qt::WindowFlags f = 0);
void hookChildrenWidgetsFocus();
protected:
bool eventFilter(QObject *object, QEvent *event);
private:
QString m_originalStyleSheet;
};
#endif // MYFRAME_H
MyFrame.cpp
#include <QEvent>
#include "MyFrame.h"
MyFrame::MyFrame(QWidget *parent, Qt::WindowFlags f)
: QFrame(parent, f)
{
m_originalStyleSheet = styleSheet();
}
void MyFrame::hookChildrenWidgetsFocus()
{
foreach (QObject *child, children()) {
if (child->isWidgetType()) {
child->installEventFilter(this);
}
}
}
bool MyFrame::eventFilter(QObject *object, QEvent *event)
{
if (event->type() == QEvent::FocusIn) {
setStyleSheet("background-color: #FFFFCC;");
} else if (event->type() == QEvent::FocusOut) {
setStyleSheet(m_originalStyleSheet);
}
return QObject::eventFilter(object, event);
}
MainWindow.cpp
#include <QHBoxLayout>
#include <QVBoxLayout>
#include <QLineEdit>
#include "MyFrame.h"
#include "mainwindow.h"
MainWindow::MainWindow(QWidget *parent) :
QMainWindow(parent)
{
setWindowTitle(tr("Test"));
MyFrame *frame1 = new MyFrame(this);
frame1->setLayout(new QVBoxLayout());
frame1->layout()->addWidget(new QLineEdit());
frame1->layout()->addWidget(new QLineEdit());
frame1->layout()->addWidget(new QLineEdit());
frame1->hookChildrenWidgetsFocus();
MyFrame *frame2 = new MyFrame(this);
frame2->setLayout(new QVBoxLayout());
frame2->layout()->addWidget(new QLineEdit());
frame2->layout()->addWidget(new QLineEdit());
frame2->layout()->addWidget(new QLineEdit());
frame2->hookChildrenWidgetsFocus();
QHBoxLayout *centralLayout = new QHBoxLayout();
centralLayout->addWidget(frame1);
centralLayout->addWidget(frame2);
QWidget *centralWidget = new QWidget();
centralWidget->setLayout(centralLayout);
setCentralWidget(centralWidget);
}
I believe the both answers you were given are wrong. They work for simple cases but are extremely fragile and clumsy. I believe that the best solution is what you actually suggested in your question. I would go for connecting to QApplication::focusChanged(from, to). You simply connect your main frame object to this signal and in the slot you check if the to object (the one which received focus) is a child of your frame object.
Frame::Frame(...)
{
// ...
connect(qApp, &QApplication::focusChanged, this, &Frame::onFocusChanged);
// ...
}
// a private method of your Frame object
void Frame::onFocusChanged(QWidget *from, QWidget *to)
{
auto w = to;
while (w != nullptr && w != this)
w = w->parentWidget();
if (w == this) // a child (or self) is focused
setStylesheet(highlightedStylesheet);
else // something else is focused
setStylesheet(normalStylesheet);
}
The advantage is obvious. This code is short and clean. You connect only one signal-slot, you do not need to catch and handle events. It responds well to any layout changes done after the object is created. And if you want to optimize away unnecessary redrawing, you should cache the information whether any child is focused and change the stylesheet only and only if this cached value gets changed. Then the solution would be prefect.
First, create a simple subclass of QFrame which reimplements the eventFilter(QObject*, QEvent*) virtual function:
class MyFrame : public QFrame {
Q_OBJECT
public:
MyFrame(QWidget *parent = 0, Qt::WindowFlags f = 0);
~MyFrame();
virtual bool eventFilter(QObject *watched, QEvent *event);
};
Use MyFrame instead of QFrame to contain your widgets. Then, somewhere in your code where you create the widgets contained in MyFrame, install an event filter on those widgets:
// ...
m_myFrame = new MyFrame(parentWidget);
QVBoxLayout *layout = new QVBoxLayout(myFrame);
m_button = new QPushButton("Widget 1", myFrame);
layout->addWidget(m_button);
m_button->installEventFilter(myFrame);
//...
At that point, MyFrame::eventFilter() will be called before any event is delivered to the widget, letting you act on it before the widget is aware of it. Within MyFrame::eventFilter(), return true if you want to filter the event out (i.e. you don't want the widget to process the event), or return false otherwise.
bool MyFrame::eventFilter(QObject *watched, QEvent *event)
{
if (watched == m_button) { // An event occured on m_button
switch (event -> type()) {
case QEvent::FocusIn:
// Change the stylesheet of the frame
break;
case QEvent::FocusOut:
// Change the stylesheet back
break;
default:
break;
}
}
return false; // We always want the event to propagate, so always return false
}

Resources